State Vector Fusion vs. Measurement Fusion for Systems with Nonlinear State Space Presentation

In this paper different sensor fusion architectures for tracking maneuvering targets are discussed. It is obvious that for tracking targets in an environment where process and measurement noises exist, applying sensor fusion approach causes the estimation results to be more accurate. But for getting the best tracking performance possible, studying different fusion architectures will be useful. The main two approaches which have been discussed here are state vector fusion and measurement fusion. For filtering and estimation issues UKF (Unscented Kalman Filter) is used as well as classical EKF (Extended Kalman filter) since these two filters have individual useful features which are helpful in different conditions. In addition to the above issues, the matter of sensor accuracies is studied, so that one understand that for the sensors which are available in each case, applying which filter algorithm with which sensor fusion algorithm will have the best performance. For stating the effect of fusion algorithm on error reduction quantitatively a new criterion entitled as fusion performance is introduced, by which the role of fusion approach is analyzed. Previous works mainly discuss linear equations for target movement and measurement model. But since these equations are nonlinear in reality, the above mentioned equations are assumed nonlinear so the problem is more close to what occurs in reality, and the analysis made is more reliable.